AI-Powered Coding: Into the Rabbit Hole? How about some Standards and Human Oversight

From TODO to WTF: When AI Coding Goes Off the Rails Without Standards

“This article was published on LinkedIn on March 21, 2025. You can see it here.”

Artificial Intelligence is revolutionizing software development. We all have heard it in the past months. From generating boilerplate code to suggesting optimizations, AI-powered tools like GitHub Copilot and OpenAI’s models are transforming how developers work. Guilty: I have been using all the tools (Cursor, Windsurf, Augmented, etc.) to try new approaches. But while AI can speed up coding, it still needs guardrails: clear standards, best practices, and human oversight to ensure quality, security, and maintainability.

Let me give you a quick example: A few days ago, I told one of these tools to build me a TODO app. Simple, right? I ended up with a huge app, fully bloated with features I was not even asking. Not only that, but when running the code, it was utterly broken. Then I asked for some fixes, and it started my way into the Rabbit Hole: the AI created new “test files” and new connections and new databases and new servers. Then, new features. And more garbage. Meanwhile, each call to the model was making someone rich. In the end, after many nonsense updates and back and forth, I got frustrated and abandoned the idea altogether.

Companies eager to adopt AI in development must recognize that standards are more critical than ever. Without well-defined guidelines, AI-generated code can introduce inconsistencies, security vulnerabilities, and accessibility issues. Developers and architects need to set the rules AI follows, ensuring that automation aligns with business goals and regulatory requirements.

That made me remember that during my time at Deloitte, I led a project focused on coding standards, ensuring that our developers followed clear coding standards and best practices (That repo is probably gold right now). At RBC, we developed a styling guide with my team, setting the foundation for consistent UI/UX across products. We had clear rules for HTML, CSS, JS, etc.

Across multiple industries, I’ve also helped with accessibility and compliance, ensuring digital experiences meet both business needs and regulatory expectations. Now more than ever, I see the value of everything I have learned along the way and all the valuable concepts the people I have worked with have taught me about the importance of clean, semantic code. Thanks to all of them.

AI will play a bigger role in coding, but it won’t replace the need for standards and governance. Organizations investing in AI-driven development must also invest in the processes and expertise required to guide it. Otherwise, they risk sacrificing long-term maintainability for short-term speed. Or, to put it in other words, they will fall into the Rabbit ole and, let me tell you, it will be costly.

The future isn’t AI vs. human developers; it’s AI and human developers working within well-defined standards to augment their knowledge. If we get this right, AI won’t just code faster; it will code smarter, more securely, and more inclusively.

How is your company preparing for AI-driven development? Are your standards ready? Are you building clear documents on how developers and AI should work? Let’s talk!